[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN111160659A - Power load prediction method considering temperature fuzzification - Google Patents

Power load prediction method considering temperature fuzzification Download PDF

Info

Publication number
CN111160659A
CN111160659A CN201911412671.8A CN201911412671A CN111160659A CN 111160659 A CN111160659 A CN 111160659A CN 201911412671 A CN201911412671 A CN 201911412671A CN 111160659 A CN111160659 A CN 111160659A
Authority
CN
China
Prior art keywords
load
data
temperature
layer
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911412671.8A
Other languages
Chinese (zh)
Other versions
CN111160659B (en
Inventor
蔡绍荣
王庆
张姝
郑瑞骁
肖先勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southwest Branch of State Grid Corp
Original Assignee
Southwest Branch of State Grid Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Southwest Branch of State Grid Corp filed Critical Southwest Branch of State Grid Corp
Priority to CN201911412671.8A priority Critical patent/CN111160659B/en
Publication of CN111160659A publication Critical patent/CN111160659A/en
Application granted granted Critical
Publication of CN111160659B publication Critical patent/CN111160659B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Human Resources & Organizations (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a power load prediction method considering temperature fuzzification, which comprises the steps of firstly, acquiring historical load data, historical temperature data and related date data of a power grid, processing the historical load data, the historical temperature data and the related date data into 15-dimensional characteristic vectors, and dividing the 15-dimensional characteristic vectors into a training data set and a testing data set according to a proportion; establishing a three-layer long-and-short term memory neural network, and performing iterative training on the three-layer long-and-short term memory neural network through a training data set to obtain a power load prediction model; and finally, inputting the forecast day data in the test data set into the power load forecasting model to obtain a power load forecast value. The method can accurately predict the short-term load by considering the time sequence characteristic of the load, and simultaneously fuzzify and process the accurate temperature to a [0,1] interval by utilizing the membership function as the characteristic vector input of the prediction model, thereby effectively reducing the random influence caused by variable and uncertain summer temperatures and improving the generalization of the power load prediction model to the temperature.

Description

Power load prediction method considering temperature fuzzification
Technical Field
The invention belongs to the technical field of power load prediction, and particularly relates to a design of a power load prediction method considering temperature fuzzification.
Background
With the development of the smart grid, an Advanced Metering Infrastructure (AMI) based on the smart meter is basically realized in the actual engineering, the smart meter can obtain 96 groups of data every day at intervals of 15mins, and a more reliable basis is provided for improving the short-term load prediction, but the actual load change can extract a change rule from the load time requirement, and the actual load is also influenced by the temperature condition to a great extent, so that the method for predicting the short-term load of the power by considering the temperature influence in summer with variable temperature has important theoretical value and practical significance.
At present, the power short-term load prediction mainly faces the following difficulties: (1) with the increase of load data quantity, the traditional time sequence prediction method has limited capability of processing nonlinear data; (2) the traditional machine learning method cannot effectively learn the data association before and after the load data with time sequence characteristics; (3) in the summer with variable temperature, the input of the traditional temperature factor may influence the prediction accuracy.
At present, a lot of researches are carried out on power load prediction by industrial scholars, and a lot of methods are proposed, but one part of related methods considers deep learning methods, the other part considers the influence of related factors such as temperature, and the influence of temperature variables on power load prediction is not considered in summer with variable temperature.
Disclosure of Invention
The invention aims to solve the problem that how to effectively fuzzify temperature variables in summer with variable temperature so as to effectively predict a power load by deep learning, and provides a power load prediction method considering temperature fuzzification.
The technical scheme of the invention is as follows: a power load prediction method considering temperature fuzzification, comprising the steps of:
and S1, collecting historical load data, historical temperature data and relevant date data of the power grid.
And S2, obtaining a 15-dimensional feature vector according to the historical load data, the historical temperature data and the related date data, and dividing the 15-dimensional feature vector into a training data set and a testing data set according to proportion.
S3, establishing a three-layer long-short time memory neural network, setting network related parameters, and performing gradient optimization on the three-layer long-short time memory neural network by using an adam algorithm.
And S4, performing iterative training on the three-layer long-short term memory neural network through a training data set, and taking the trained three-layer long-short term memory neural network as a power load prediction model.
And S5, inputting the forecast day data in the test data set into the power load forecasting model to obtain a power load forecast value.
Further, step S2 includes the following substeps:
and S21, carrying out normalization processing on the historical load data to obtain normalized historical load data.
And S22, obtaining 8-dimensional related load characteristic vectors according to the normalized historical load data.
And S23, performing fuzzification processing on the historical temperature data to obtain a 6-dimensional fuzzy temperature characteristic vector.
And S24, obtaining a 1-dimensional date attribute feature vector according to the related date data.
And S25, combining the 8-dimensional related load characteristic vector, the 6-dimensional fuzzy temperature characteristic vector and the 1-dimensional date attribute characteristic vector into a 15-dimensional characteristic vector, and dividing the 15-dimensional characteristic vector into a training data set and a testing data set according to the ratio of 8: 2.
Further, the formula for performing the normalization process on the historical load data in step S21 is:
Figure BDA0002350386590000021
wherein Y represents normalized historical load data, YmaxIs a fixed value, and takes the value of 1, YminIs a fixed value, takes a value of 0, X represents a load value, XmaxAnd XminRespectively representing the maximum and minimum values in the historical load data.
Further, step S22 is specifically: selecting the load L at the time t from the normalized historical load datatBy a load LtLoad L at the previous momentt-1Load L at the next momentt+1Load at time t of day before
Figure BDA0002350386590000022
Load at time t-1 of previous day
Figure BDA0002350386590000023
Load at time t +1 of previous day
Figure BDA0002350386590000024
Load at time t of previous week
Figure BDA0002350386590000025
Load at time t-1 of previous week
Figure BDA0002350386590000026
And the load at the time of the previous week t +1
Figure BDA0002350386590000027
Construction of 8-dimensional related load feature vectors
Figure BDA0002350386590000028
Further, step S23 is specifically: selecting temperature T at T moment from historical temperature datatRespectively inputting the low-temperature day membership function, the medium-temperature day membership function and the high-temperature day membership function to obtain TtFuzzified feature vectors of
Figure BDA0002350386590000029
Average temperature T of the day at the moment TavrgRespectively inputting a low-temperature day membership function, a medium-temperature day membership function and a high-temperature day membership function to obtain TavrgFuzzified feature vectors of
Figure BDA00023503865900000210
Will TtFuzzified feature vectors of
Figure BDA00023503865900000211
And to TavrgFuzzified feature vectors of
Figure BDA00023503865900000212
Combining to obtain 6-dimensional fuzzy temperature characteristic vector
Figure BDA00023503865900000213
Further, a low temperature day membership function m1The calculation formula of (2) is as follows:
Figure BDA0002350386590000031
medium temperature day membership function m2The calculation formula of (2) is as follows:
Figure BDA0002350386590000032
high temperature day membership function m3The calculation formula of (2) is as follows:
Figure BDA0002350386590000033
where T represents the input temperature.
Further, step S24 is specifically: according to the related date data, the date attribute D of the working day is dividedw/rSet to 1, date attribute D of holidayw/rSet to 0, get 1-dimensional date attribute feature vector [ Dw/r]。
Further, the three-layer long-short term memory neural network established in step S3 includes 3 hidden layers and a full-connection layer which are connected in sequence, the neuron number of each hidden layer is set to 25, the optimizers are Adam, the root mean square error RMSE is selected as a loss function of the three-layer long-short term memory neural network, and the maximum iteration number of the network is set to 500; every hides layer structure the same, all includes forgetting gate, input gate and the output gate that connects gradually.
The calculation formula of the forgetting door is as follows:
ft＝σ(Wf[Yt-1,Xt]+bf)
wherein f istRepresents the output of the forgetting gate at time t, (. sigma.) represents the sigmoid function, WfMatrix weight of forgetting gate, Yt-1Hidden layer output for time t-1, XtHidden layer input for time t, bfIs the bias term for the forgetting gate.
The input gate includes two sections: firstly, the useful information input is updated through a sigmoid layer, and then a new candidate value state vector is created through a tanh layer, wherein the calculation formula is as follows:
ut＝σ(Wu[Yt-1,Xt]+bu)
ht＝tanh(Wh[Yt-1,Xt]+bh)
wherein u istIndicating the state of the input gate at time t, WuMatrix weights representing the sigmoid layer of the input gate, buOffset term, h, representing the sigmoid layer of the input gatetState vector of newly added candidate value, W, representing time thMatrix weights representing the input gate tanh layer, bhIndicating the bias term of the input gate tanh layer.
The output gate includes two parts: firstly, filtering information to be output through a sigmoid layer, and then hiding a layer state CtPassing through a tanh layer to make the state range of [ -1,1 [)]Multiplying the output of the sigmoid layer and the output of the tanh layer to obtain the expected output of the hidden layer, wherein the calculation formula is as follows:
ot＝σ(Wo[Yt-1,Xt]+bo)
Yt＝ot*tanh(Ct)
wherein o istIndicating the output gate state at time t, WoMatrix weights representing sigmoid layers of the output gates, boBias term, Y, representing the sigmoid layer of the output gatetRepresenting the expected output of the hidden layer at time t, CtThe hidden layer state at the time t is represented by the following calculation formula:
Ct＝ft*Ct-1+ut*ht
wherein C ist-1Representing the hidden layer state at time t-1.
Further, step S4 includes the following substeps:
and S41, inputting the training set data into a three-layer long-and-short time memory neural network to obtain a predicted value, and calculating an error through a loss function.
And S42, optimizing back propagation through an Adam algorithm, and updating network related parameters.
And S43, judging whether the boundary condition or the maximum iteration number of the network related parameters is reached, if so, obtaining the trained three-layer long-and-short time memory neural network, taking the three-layer long-and-short time memory neural network as a power load prediction model, and entering the step S5, otherwise, returning to the step S41.
Further, the formula of the Adam algorithm in step S42 in the t-th iteration is:
Figure BDA0002350386590000051
wherein M isdw、MdbFor the parameter calculated by the Momentum algorithm, Rdw、RdbThe parameters calculated for the RMSProp algorithm,
Figure BDA0002350386590000052
and
Figure BDA0002350386590000053
are all made ofThe offset correction made by the Adam algorithm,
Figure BDA0002350386590000054
is the gradient index in the Momentum algorithm,
Figure BDA0002350386590000055
is the gradient index in the RMSProp algorithm, epsilon is the coefficient of the smoothing term, Wt、Wt-1Representing the matrix weights of the sigmoid layers at time t and time t-1, respectively, bt、bt-1Bias terms representing sigmoid layers at time t and time t-1, respectively, are learning rates α.
The invention has the beneficial effects that: according to the three-layer long-short time memory neural network based on Adam optimization, the short-term load can be accurately predicted by considering the load time sequence characteristics, and meanwhile, the accurate temperature is fuzzified to a [0,1] interval by using a membership function and is used as the characteristic vector input of a prediction model, so that the randomness influence caused by variable and uncertain summer temperatures can be effectively reduced, and the temperature generalization of a short-term load prediction model is improved.
Drawings
Fig. 1 is a flowchart illustrating a power load prediction method considering temperature fuzzification according to an embodiment of the present invention.
FIG. 2 is a schematic diagram of a temperature membership function according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of a three-layer long-short term memory neural network according to an embodiment of the present invention.
Fig. 4 is a schematic view of a hidden layer structure according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It is to be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, not to limit the scope of the invention.
The embodiment of the invention provides a power load prediction method considering temperature fuzzification, as shown in fig. 1, the method comprises the following steps of S1-S5:
and S1, collecting historical load data, historical temperature data and relevant date data of the power grid.
In the embodiment of the invention, summer regional load data with a sampling point of 96(15mins per point) on one day is selected as the historical load data, temperature data with a sampling point of 96 is selected as the historical temperature data, and the collection of related date data is that the collection day is a working day or a rest day.
And S2, obtaining a 15-dimensional feature vector according to the historical load data, the historical temperature data and the related date data, and dividing the 15-dimensional feature vector into a training data set and a testing data set according to proportion.
The step S2 includes the following substeps S21-S25:
and S21, carrying out normalization processing on the historical load data to obtain normalized historical load data.
In the embodiment of the present invention, the formula for performing normalization processing on historical load data is as follows:
Figure BDA0002350386590000061
wherein Y represents normalized historical load data, YmaxIs a fixed value, and takes the value of 1, YminIs a fixed value, takes a value of 0, X represents a load value, XmaxAnd XminRespectively representing the maximum and minimum values in the historical load data.
And S22, obtaining 8-dimensional related load characteristic vectors according to the normalized historical load data.
Considering the load data itself, assume that the load at time t is predicted to be LtSince the load curve is continuous and non-abrupt, the load sum L at a time before and after time ttHas stronger correlation; meanwhile, the load curves have the characteristics of daily periodicity and weekly periodicity, and the loads at the time of t-1, t and t +1 of the previous day and the previous week are also considered.
Therefore, in the embodiment of the invention, the load at the time t is selected from the normalized historical load data to be LtBy a load LtLoad L at the previous momentt-1The latter oneLoad at time Lt+1Load at time t of day before
Figure BDA0002350386590000062
Load at time t-1 of previous day
Figure BDA0002350386590000063
Load at time t +1 of previous day
Figure BDA0002350386590000064
Load at time t of previous week
Figure BDA0002350386590000065
Load at time t-1 of previous week
Figure BDA0002350386590000066
And the load at the time of the previous week t +1
Figure BDA0002350386590000067
Construction of 8-dimensional related load feature vectors
Figure BDA0002350386590000068
And S23, performing fuzzification processing on the historical temperature data to obtain a 6-dimensional fuzzy temperature characteristic vector.
In the embodiment of the invention, an isosceles triangle membership function is constructed for historical temperature data, and the daily average temperature T is usedavrgAnd the temperature T at time TtAs the input of fuzzy variables, the fuzzy processing of the history temperature data is realized, and the specific method comprises the following steps:
selecting temperature T at T moment from historical temperature datatRespectively inputting the low-temperature day membership function, the medium-temperature day membership function and the high-temperature day membership function to obtain TtFuzzified feature vectors of
Figure BDA0002350386590000069
Average temperature T of the day at the moment TavrgRespectively inputting a low-temperature day membership function and a medium-temperature day membership functionAnd the high-temperature day membership function to obtain TavrgFuzzified feature vectors of
Figure BDA0002350386590000071
Fuzzifying feature vectors of T
Figure BDA0002350386590000072
And to TavrgFuzzified feature vectors of
Figure BDA0002350386590000073
Combining to obtain 6-dimensional fuzzy temperature characteristic vector
Figure BDA0002350386590000074
In the embodiment of the invention, the temperature is divided into low temperature days (20-26 ℃), medium temperature days (23-32 ℃) and high temperature days (29-35 ℃), wherein the low temperature day membership function m1The calculation formula of (2) is as follows:
Figure BDA0002350386590000075
medium temperature day membership function m2The calculation formula of (2) is as follows:
Figure BDA0002350386590000076
high temperature day membership function m3The calculation formula of (2) is as follows:
Figure BDA0002350386590000077
where T represents the input temperature.
A schematic of the membership function is shown in FIG. 2, where the shaded area represents the temperature fuzzification area.
And S24, obtaining a 1-dimensional date attribute feature vector according to the related date data.
In the embodiment of the invention, the date attribute D of the working day is converted according to the related date dataw/rSet to 1, date attribute D of holidayw/rSet to 0, get 1-dimensional date attribute feature vector [ Dw/r]。
S25, combining the 8-dimensional related load feature vector, the 6-dimensional fuzzy temperature feature vector and the 1-dimensional date attribute feature vector into a 15-dimensional feature vector
Figure BDA0002350386590000078
And divided into a training data set and a test data set in an 8:2 ratio.
S3, establishing a three-layer long-short time memory neural network, setting network related parameters, and performing gradient optimization on the three-layer long-short time memory neural network by using an adam algorithm.
In the embodiment of the invention, a three-layer long-short time memory neural network model is established by using an lstmayer () function in matlab, as shown in fig. 3, the three-layer long-short time memory neural network comprises 3 hidden layers and a full-connection layer which are sequentially connected, the number of neurons in each hidden layer is set to be 25, optimizers are Adam, a root mean square error RMSE is selected as a loss function (Lose function) of the three-layer long-short time memory neural network, and the maximum iteration number of the network is set to be 500.
As shown in fig. 4, each hidden layer has the same structure and includes a forgetting gate, an input gate, and an output gate, which are connected in sequence.
Wherein, forget the door and can carry out selectivity forgetting to information, its computational formula is:
ft＝σ(Wf[Yt-1,Xt]+bf)
wherein f istRepresents the output of the forgetting gate at time t, (. sigma.) represents the sigmoid function, WfMatrix weight of forgetting gate, Yt-1Hidden layer output for time t-1, XtHidden layer input for time t, bfIs the bias term for the forgetting gate.
The input gate includes two sections: firstly, the useful information input is updated through a sigmoid layer, and then a new candidate value state vector is created through a tanh layer, wherein the calculation formula is as follows:
ut＝σ(Wu[Yt-1,Xt]+bu)
ht＝tanh(Wh[Yt-1,Xt]+bh)
wherein u istIndicating the state of the input gate at time t, WuMatrix weights representing the sigmoid layer of the input gate, buOffset term, h, representing the sigmoid layer of the input gatetState vector of newly added candidate value, W, representing time thMatrix weights representing the input gate tanh layer, bhIndicating the bias term of the input gate tanh layer.
The output gate includes two parts: firstly, filtering information to be output through a sigmoid layer, and then hiding a layer state CtPassing through a tanh layer to make the state range of [ -1,1 [)]Multiplying the output of the sigmoid layer and the output of the tanh layer to obtain the expected output of the hidden layer, wherein the calculation formula is as follows:
ot＝σ(Wo[Yt-1,Xt]+bo)
Yt＝ot*tanh(Ct)
wherein o istIndicating the output gate state at time t, WoMatrix weights representing sigmoid layers of the output gates, boBias term, Y, representing the sigmoid layer of the output gatetRepresenting the expected output of the hidden layer at time t, CtThe hidden layer state at time t is represented, which is determined by adding the information part controlled by the forgetting gate and the input gate: long time information composed oftControlling; short time information is composed oftAnd htAnd (3) controlling, wherein the calculation formula is as follows:
Ct＝ft*Ct-1+ut*ht
wherein C ist-1Representing the hidden layer state at time t-1.
And S4, performing iterative training on the three-layer long-short term memory neural network through a training data set, and taking the trained three-layer long-short term memory neural network as a power load prediction model.
The step S4 includes the following substeps S41-S43:
and S41, inputting the training set data into a three-layer long-and-short time memory neural network to obtain a predicted value, and calculating an error through a loss function.
And S42, optimizing back propagation through an Adam algorithm, and updating network related parameters.
In the embodiment of the invention, the formula of the Adam algorithm in the t-th iteration is as follows:
Figure BDA0002350386590000091
wherein M isdw、MdbFor the parameter calculated by the Momentum algorithm, Rdw、RdbThe parameters calculated for the RMSProp algorithm,
Figure BDA0002350386590000092
and
Figure BDA0002350386590000093
are all deviation corrections made by the Adam algorithm,
Figure BDA0002350386590000094
is the gradient index in the Momentum algorithm, generally takes the value of 0.9,
Figure BDA0002350386590000095
the gradient index in the RMSProp algorithm is generally 0.999, epsilon is a smooth term coefficient and is generally 10-8，Wt、Wt-1Representing the matrix weights of the sigmoid layers at time t and time t-1, respectively, bt、bt-1Bias terms representing sigmoid layers at time t and time t-1, respectively, are learning rates α.
And S43, judging whether the boundary condition or the maximum iteration number of the network related parameters is reached, if so, obtaining the trained three-layer long-and-short time memory neural network, taking the three-layer long-and-short time memory neural network as a power load prediction model, and entering the step S5, otherwise, returning to the step S41.
And S5, inputting the forecast day data in the test data set into the power load forecasting model to obtain a power load forecast value.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims (10)

1. A power load prediction method considering temperature fuzzification is characterized by comprising the following steps:
s1, collecting historical load data, historical temperature data and relevant date data of the power grid;
s2, obtaining a 15-dimensional feature vector according to the historical load data, the historical temperature data and the related date data, and dividing the 15-dimensional feature vector into a training data set and a testing data set according to the proportion;
s3, establishing a three-layer long-short time memory neural network, setting network related parameters, and performing gradient optimization on the three-layer long-short time memory neural network by using an adam algorithm;
s4, carrying out iterative training on the three-layer long-short term memory neural network through a training data set, and taking the trained three-layer long-short term memory neural network as a power load prediction model;
and S5, inputting the forecast day data in the test data set into the power load forecasting model to obtain a power load forecast value.
2. The power load prediction method according to claim 1, wherein the step S2 includes the following substeps:
s21, carrying out normalization processing on the historical load data to obtain normalized historical load data;
s22, obtaining 8-dimensional related load characteristic vectors according to the normalized historical load data;
s23, fuzzifying the historical temperature data to obtain a 6-dimensional fuzzy temperature characteristic vector;
s24, obtaining a 1-dimensional date attribute feature vector according to the related date data;
and S25, combining the 8-dimensional related load characteristic vector, the 6-dimensional fuzzy temperature characteristic vector and the 1-dimensional date attribute characteristic vector into a 15-dimensional characteristic vector, and dividing the 15-dimensional characteristic vector into a training data set and a testing data set according to the ratio of 8: 2.
3. The power load prediction method according to claim 2, wherein the formula for normalizing the historical load data in step S21 is:
Figure FDA0002350386580000011
wherein Y represents normalized historical load data, YmaxIs a fixed value, and takes the value of 1, YminIs a fixed value, takes a value of 0, X represents a load value, XmaxAnd XminRespectively representing the maximum and minimum values in the historical load data.
4. The power load prediction method according to claim 2, wherein the step S22 is specifically: selecting the load L at the time t from the normalized historical load datatBy a load LtLoad L at the previous momentt-1Load L at the next momentt+1Load at time t of day before
Figure FDA0002350386580000012
Load at time t-1 of previous day
Figure FDA0002350386580000013
Load at time t +1 of previous day
Figure FDA0002350386580000014
Load at time t of previous week
Figure FDA0002350386580000015
Load at time t-1 of previous week
Figure FDA0002350386580000016
And the load at the time of the previous week t +1
Figure FDA0002350386580000017
Construction of 8-dimensional related load feature vectors
Figure FDA0002350386580000018
5. The power load prediction method according to claim 2, wherein the step S23 is specifically: selecting temperature T at T moment from historical temperature datatRespectively inputting the low-temperature day membership function, the medium-temperature day membership function and the high-temperature day membership function to obtain TtFuzzified feature vectors of
Figure FDA0002350386580000021
Average temperature T of the day at the moment TavrgRespectively inputting a low-temperature day membership function, a medium-temperature day membership function and a high-temperature day membership function to obtain TavrgFuzzified feature vectors of
Figure FDA0002350386580000022
Will TtFuzzified feature vectors of
Figure FDA0002350386580000023
And to TavrgFuzzified feature vectors of
Figure FDA0002350386580000024
Combining to obtain 6-dimensional fuzzy temperature characteristic vector
Figure FDA0002350386580000025
6. According to claimThe method for predicting an electric power load according to claim 5, wherein the low-temperature daily membership function m1The calculation formula of (2) is as follows:
Figure FDA0002350386580000026
the medium temperature daily membership function m2The calculation formula of (2) is as follows:
Figure FDA0002350386580000027
the high temperature day membership function m3The calculation formula of (2) is as follows:
Figure FDA0002350386580000028
where T represents the input temperature.
7. The power load prediction method according to claim 2, wherein the step S24 is specifically: according to the related date data, the date attribute D of the working day is dividedw/rSet to 1, date attribute D of holidayw/rSet to 0, get 1-dimensional date attribute feature vector [ Dw/r]。
8. The power load prediction method according to claim 1, wherein the three-layer long-short term memory neural network established in step S3 includes 3 hidden layers and a fully-connected layer which are connected in sequence, the number of neurons in each hidden layer is set to 25, the optimizers are Adam, the root mean square error RMSE is selected as a loss function of the three-layer long-short term memory neural network, and the maximum number of iterations of the network is set to 500; each hidden layer has the same structure and comprises a forgetting gate, an input gate and an output gate which are connected in sequence;
the calculation formula of the forgetting door is as follows:
ft＝σ(Wf[Yt-1,Xt]+bf)
wherein f istRepresents the output of the forgetting gate at time t, (. sigma.) represents the sigmoid function, WfMatrix weight of forgetting gate, Yt-1Hidden layer output for time t-1, XtHidden layer input for time t, bfA bias term for a forget gate;
the input gate includes two parts: firstly, the useful information input is updated through a sigmoid layer, and then a new candidate value state vector is created through a tanh layer, wherein the calculation formula is as follows:
ut＝σ(Wu[Yt-1,Xt]+bu)
ht＝tanh(Wh[Yt-1,Xt]+bh)
wherein u istIndicating the state of the input gate at time t, WuMatrix weights representing the sigmoid layer of the input gate, buOffset term, h, representing the sigmoid layer of the input gatetState vector of newly added candidate value, W, representing time thMatrix weights representing the input gate tanh layer, bhAn offset term representing the input gate tanh layer;
the output gate includes two portions: firstly, filtering information to be output through a sigmoid layer, and then hiding a layer state CtPassing through a tanh layer to make the state range of [ -1,1 [)]Multiplying the output of the sigmoid layer and the output of the tanh layer to obtain the expected output of the hidden layer, wherein the calculation formula is as follows:
ot＝σ(Wo[Yt-1,Xt]+bo)
Yt＝ot*tanh(Ct)
wherein o istIndicating the output gate state at time t, WoMatrix weights representing sigmoid layers of the output gates, boBias term, Y, representing the sigmoid layer of the output gatetRepresenting the expected output of the hidden layer at time t, CtThe hidden layer state at the time t is represented by the following calculation formula:
Ct＝ft*Ct-1+ut*ht
wherein C ist-1Representing the hidden layer state at time t-1.
9. The power load prediction method according to claim 8, wherein the step S4 includes the following substeps:
s41, inputting the training set data into a three-layer long-and-short time memory neural network to obtain a predicted value, and calculating an error through a loss function;
s42, optimizing back propagation through an Adam algorithm, and updating network related parameters;
and S43, judging whether the boundary condition or the maximum iteration number of the network related parameters is reached, if so, obtaining the trained three-layer long-and-short time memory neural network, taking the three-layer long-and-short time memory neural network as a power load prediction model, and entering the step S5, otherwise, returning to the step S41.
10. The method according to claim 9, wherein the Adam algorithm in step S42 has the formula in the t-th iteration:
Figure FDA0002350386580000041
wherein M isdw、MdbFor the parameter calculated by the Momentum algorithm, Rdw、RdbThe parameters calculated for the RMSProp algorithm,
Figure FDA0002350386580000042
and
Figure FDA0002350386580000043
are all deviation corrections made by the Adam algorithm,
Figure FDA0002350386580000044
is the gradient index in the Momentum algorithm,
Figure FDA0002350386580000045
for gradients in the RMSProp algorithmExponent, ε is the coefficient of the smoothing term, Wt、Wt-1Representing the matrix weights of the sigmoid layers at time t and time t-1, respectively, bt、bt-1Bias terms representing sigmoid layers at time t and time t-1, respectively, are learning rates α.
CN201911412671.8A 2019-12-31 2019-12-31 Power load prediction method considering temperature fuzzification Active CN111160659B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911412671.8A CN111160659B (en) 2019-12-31 2019-12-31 Power load prediction method considering temperature fuzzification

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911412671.8A CN111160659B (en) 2019-12-31 2019-12-31 Power load prediction method considering temperature fuzzification

Publications (2)

Publication Number Publication Date
CN111160659A true CN111160659A (en) 2020-05-15
CN111160659B CN111160659B (en) 2021-11-23

Family

ID=70560034

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911412671.8A Active CN111160659B (en) 2019-12-31 2019-12-31 Power load prediction method considering temperature fuzzification

Country Status (1)

Country Link
CN (1) CN111160659B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112200391A (en) * 2020-11-17 2021-01-08 国网陕西省电力公司经济技术研究院 Load forecasting method at edge side of distribution network based on k-nearest neighbor mutual information feature simplification
CN112329983A (en) * 2020-09-30 2021-02-05 联想(北京)有限公司 Data processing method and device
CN112418560A (en) * 2020-12-10 2021-02-26 长春理工大学 PM2.5 concentration prediction method and system
CN112685900A (en) * 2020-12-31 2021-04-20 国网浙江省电力有限公司营销服务中心 Power load simulation method for representing impact load power characteristics
CN115809610A (en) * 2023-02-08 2023-03-17 广东电网有限责任公司中山供电局 Direct-buried three-core cable current-carrying capacity prediction method and system based on multi-step load

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107239859A (en) * 2017-06-05 2017-10-10 国网山东省电力公司电力科学研究院 The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term
CN108830487A (en) * 2018-06-21 2018-11-16 王芊霖 Methods of electric load forecasting based on long neural network in short-term
CN109255505A (en) * 2018-11-20 2019-01-22 国网辽宁省电力有限公司经济技术研究院 A kind of short-term load forecasting method of multi-model fused neural network
CN109685265A (en) * 2018-12-21 2019-04-26 积成电子股份有限公司 A kind of prediction technique of power-system short-term electric load
CN110298501A (en) * 2019-06-21 2019-10-01 河海大学常州校区 Electric load prediction technique based on long Memory Neural Networks in short-term

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107239859A (en) * 2017-06-05 2017-10-10 国网山东省电力公司电力科学研究院 The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term
CN108830487A (en) * 2018-06-21 2018-11-16 王芊霖 Methods of electric load forecasting based on long neural network in short-term
CN109255505A (en) * 2018-11-20 2019-01-22 国网辽宁省电力有限公司经济技术研究院 A kind of short-term load forecasting method of multi-model fused neural network
CN109685265A (en) * 2018-12-21 2019-04-26 积成电子股份有限公司 A kind of prediction technique of power-system short-term electric load
CN110298501A (en) * 2019-06-21 2019-10-01 河海大学常州校区 Electric load prediction technique based on long Memory Neural Networks in short-term

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
DIEDERIK P. KINGMA等: ""ADAM: A METHOD FOR STOCHASTIC OPTIMIZATION"", 《ICLR 2015》 *
王小新: ""一文看懂各种神经网络优化算法：从梯度下降到Adam方法"", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/27449596》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112329983A (en) * 2020-09-30 2021-02-05 联想(北京)有限公司 Data processing method and device
CN112329983B (en) * 2020-09-30 2024-07-26 联想(北京)有限公司 Data processing method and device
CN112200391A (en) * 2020-11-17 2021-01-08 国网陕西省电力公司经济技术研究院 Load forecasting method at edge side of distribution network based on k-nearest neighbor mutual information feature simplification
CN112200391B (en) * 2020-11-17 2023-07-04 国网陕西省电力公司经济技术研究院 Power distribution network edge side load prediction method based on k-nearest neighbor mutual information feature simplification
CN112418560A (en) * 2020-12-10 2021-02-26 长春理工大学 PM2.5 concentration prediction method and system
CN112418560B (en) * 2020-12-10 2024-05-14 长春理工大学 PM2.5 concentration prediction method and system
CN112685900A (en) * 2020-12-31 2021-04-20 国网浙江省电力有限公司营销服务中心 Power load simulation method for representing impact load power characteristics
CN112685900B (en) * 2020-12-31 2023-09-26 国网浙江省电力有限公司营销服务中心 An electric load simulation method to characterize the power characteristics of impact loads
CN115809610A (en) * 2023-02-08 2023-03-17 广东电网有限责任公司中山供电局 Direct-buried three-core cable current-carrying capacity prediction method and system based on multi-step load

Also Published As

Publication number Publication date
CN111160659B (en) 2021-11-23

Similar Documents

Publication Publication Date Title
CN111160659B (en) Power load prediction method considering temperature fuzzification
Liang et al. A novel wind speed prediction strategy based on Bi-LSTM, MOOFADA and transfer learning for centralized control centers
CN114626512B (en) A high temperature disaster prediction method based on directed graph neural network
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
CN109685252B (en) Building energy consumption prediction method based on cyclic neural network and multi-task learning model
CN112949945A (en) Wind power ultra-short-term prediction method for improving bidirectional long-short term memory network
CN111967688B (en) Power load prediction method based on Kalman filter and convolutional neural network
CN108280551A (en) A kind of photovoltaic power generation power prediction method using shot and long term memory network
CN113554466B (en) Short-term electricity consumption prediction model construction method, prediction method and device
CN112116147A (en) River water temperature prediction method based on LSTM deep learning
CN109583565B (en) Flood prediction method based on attention model long-time and short-time memory network
CN115099519B (en) Oil well yield prediction method based on multi-machine learning model fusion
CN112036658B (en) Wind power real-time prediction method based on forgetting long-term and short-term memory with difference
CN108764539A (en) A kind of water levels of upstream and downstream prediction technique of step hydropower station
CN111985719B (en) Power load prediction method based on improved long-term and short-term memory network
CN114862032B (en) XGBoost-LSTM-based power grid load prediction method and device
CN110212524A (en) A kind of region Methods of electric load forecasting
CN114492922A (en) Medium-and-long-term power generation capacity prediction method
CN116345555A (en) A Short-Term Photovoltaic Power Forecasting Method Based on CNN-ISCA-LSTM Model
CN115759415A (en) Electricity Demand Forecasting Method Based on LSTM-SVR
CN114372631A (en) A runoff prediction method in data-deficient areas based on small sample learning and LSTM
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
Zhichao et al. Short-term load forecasting of multi-layer LSTM neural network considering temperature fuzzification
CN116681159A (en) Short-term power load prediction method based on whale optimization algorithm and DRESN
CN113159395A (en) Deep learning-based sewage treatment plant water inflow prediction method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Cai Shaorong

Inventor after: Wei Mingkui

Inventor after: Zhou Quan

Inventor after: Lu Liang

Inventor after: Jiang Li

Inventor after: Wang Qing

Inventor after: Zheng Ruixiao

Inventor after: Zhang Shu

Inventor after: Xiao Xianyong

Inventor before: Cai Shaorong

Inventor before: Wang Qing

Inventor before: Zhang Shu

Inventor before: Zheng Ruixiao

Inventor before: Xiao Xianyong

GR01 Patent grant
GR01 Patent grant